Load prediction method in multi-extreme weather scene considering error correction

Abstract

The invention discloses a load prediction method in a multi-extreme weather scene considering error correction. The method comprises the following steps: a correlation analysis and screening system of various meteorological factors and extreme weather; a time sequence feature extraction and denoising processing system of the load time sequence; by introducing and establishing a prediction error objective function and performing hyper-parameter optimization on the BiGRU through a Bayesian method, the power load prediction precision under an extreme weather condition is remarkably improved, so that a powerful guarantee is provided for stable operation of a power grid. The ICEEMDAN algorithm and the VMD algorithm are adopted to perform secondary decomposition and reconstruction of the time sequence, and the deep processing technology can more finely capture complex local features and seasonal patterns in the data. Through Pearson correlation analysis, extreme weather characteristics closely related to load prediction can be effectively identified. The feature recognition technology enables the model to be more focused on meteorological factors having significant influence on load change, and the prediction capability of the model in extreme weather is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of accurate prediction of electric load at multiple time and space scales, and in particular to a method for load prediction under multiple extreme weather scenarios taking error correction into account. Background Art

[0002] At present, there have been a large number of research results on the problem of power load forecasting, mainly including relevant methods based on traditional mathematical statistics and relevant machine learning methods based on data-driven.

[0003] Under extreme weather conditions, power load has stronger volatility and mutation. Existing forecasting methods are difficult to capture the accurate change pattern of load, thus affecting the prediction accuracy. To solve this problem, this paper proposes a load forecasting method under multiple extreme weather scenarios taking into account error correction, which can effectively deal with the mutation, non-stationarity and strong randomness of power load time series under extreme weather. It aims to enhance the adaptability of the power system to weather changes through more accurate load forecasting, ensure the stability and economy of power supply, and meet the needs of modern power systems for efficient management under diversified energy structures. Summary of the invention

[0004] In view of the above problems, the present invention is proposed

[0005] Therefore, the technical problem solved by the present invention is: Therefore, the existing related methods based on traditional mathematical statistics and the related machine learning methods based on data-driven have the problem of difficulty in capturing the accurate changing patterns of the load, which affects the prediction accuracy. The technical problem solved by the present invention is: the existing problem of how to reduce the impact of the mutation, non-stationarity and strong randomness of the power load time series under extreme weather on the prediction accuracy.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a load forecasting method under multiple extreme weather scenarios taking into account error correction, comprising:

[0007] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, wherein:

[0008] The correlation between extreme weather classification and meteorological indicators was analyzed through the Pearson correlation coefficient;

[0009] According to the meteorological index, an extreme weather classification corresponding to the meteorological index is obtained, and a comprehensive extreme weather evaluation system considering the meteorological index is established;

[0010] Use two channels to obtain the initial parameters for load forecasting; use channel one to collect a clear historical meteorological data set, and use channel two to obtain an artificial test simulated meteorological data set;

[0011] Taking extreme weather classification and meteorological indicator datasets as input, the temporal features of the time series are decomposed, denoised, and reconstructed;

[0012] The results of time series feature extraction are input into the main load prediction model and error correction model to obtain the power load prediction results.

[0013] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, the Pearson correlation coefficient includes obtaining the main meteorological indicators affecting each extreme weather by screening the meteorological characteristic factors affecting extreme weather and analyzing the relationship between specific weather classifications and meteorological indicators.

[0014] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, wherein: the comprehensive evaluation includes constructing an extreme weather comprehensive evaluation index model by constructing different weights and input parameters;

[0015] R H =κ h x1+κ y x2+...+κ d x n

[0016] Among them, R H is the evaluation value of extreme weather type, x1, x2...x n is the meteorological index of extreme weather type, and the meteorological index weight is κ h , κ d ... y , are all greater than 0 and the sum is 0;

[0017] The specific functional relationship between extreme weather and meteorological indicators is reflected through the extreme weather comprehensive evaluation index model.

[0018] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, the artificial test instruction simulation includes: inputting a specific weather classification, obtaining the meteorological index corresponding to the input weather classification according to the Pearson correlation coefficient, combining the meteorological indexes corresponding to the weather classification according to the calculation formula of the extreme weather comprehensive evaluation index, and simulating various extreme weather conditions;

[0019] In the first channel, meteorological indicators are input and brought into the comprehensive evaluation indicator model of each extreme weather classification through the matching mechanism to obtain the evaluation value for each extreme weather type. Through the correspondence between the evaluation value of each extreme weather and the threshold, the specific extreme weather classification is matched;

[0020] In the second channel, the extreme weather type is input, and a specific comprehensive evaluation index model is obtained based on the input extreme weather type. According to the threshold range of the evaluation value, the meteorological indicators in the comprehensive evaluation index model are assigned and randomly combined to obtain the meteorological indicators corresponding to the weather type.

[0021] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, wherein: the load time series includes secondary decomposition and reconstruction of the load time series to form a highly denoised and feature-clear time series;

[0022] The specific steps are:

[0023] The ICEEMDAN algorithm is used to decompose the original multivariate time series load data;

[0024] The algorithm ICEEMDAN is used to decompose complex time series data into a series of intrinsic mode functions with different time scales;

[0025] The K-means clustering algorithm is used to classify the intrinsic mode function of the sample entropy; the four steps of distance measurement, cluster allocation, cluster center update, and objective function are respectively represented;

[0026] Extracting high-frequency components of the intrinsic mode function according to its complexity;

[0027] Applying variational mode decomposition algorithm to perform secondary decomposition on the eigenmode function belonging to the high frequency part; using the calculation of variational mode decomposition algorithm to decompose the signal into eigenmode functions with specific frequency characteristics;

[0028] A threshold is selected according to a wavelet threshold denoising method; and components identified as containing noise are processed using the wavelet packet decomposition, threshold quantization and signal reconstruction techniques.

[0029] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, the decomposition, denoising and reconstruction include combining and reconstructing subsequences through clear time series features after process processing; processing sequence data through two gated recurrent units, so that the bidirectional gated recurrent unit neural network model simultaneously obtains the context information of each time point, and predicts the changes in power load under extreme weather conditions that are jointly affected by historical and future trends.

[0030] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, wherein: the input load main forecasting model and the error correction model include using a normalized exponential function to further improve the model forecasting effect; the forecast error objective function f is used to evaluate the performance of the model, and the model parameters are optimized by minimizing this function;

[0031] Due to the time series characteristics and nonlinear features, as well as the complexity of hyperparameter selection of the CNN-BiGRU model, the Bayesian optimization method is used to dynamically adjust the search strategy for the existing evaluation results, identify the hyperparameter configuration, avoid falling into the local optimal solution, and achieve error correction.

[0032] As a preferred solution of the load forecasting method under multiple extreme weather scenarios taking into account error correction described in the present invention, the prediction error objective function f is a mathematical expression that measures the difference between the model prediction value and the actual observation value; the error function is used to evaluate the performance of the model and optimize the model parameters by minimizing this function.

[0033] As a preferred solution of the method for power load forecasting under multiple extreme weather scenarios taking into account error correction described in the present invention, it includes: a classification module, an analysis and simulation module, and a prediction module;

[0034] The classification module is used to: analyze the correlation between extreme weather classification and the meteorological index through the Pearson correlation coefficient; obtain the extreme weather classification corresponding to the meteorological index according to the meteorological index, and establish a comprehensive evaluation system for extreme weather taking into account the meteorological index;

[0035] The analysis and simulation module is used to: use two channels to obtain initial parameters for load forecasting; collect clear historical meteorological data sets through channel one, and obtain artificial test simulation meteorological data sets through channel two;

[0036] The prediction module is used to: take the classification of extreme weather and meteorological index data sets as input, decompose, denoise and reconstruct the temporal characteristics of the time series; input the results of the temporal characteristics extraction into the load main prediction model and the error correction model to obtain the power load prediction results.

[0037] A computer device comprises a memory and a processor. The memory stores steps of a load forecasting method under multiple extreme weather scenarios taking error correction into account when a computer program is executed by the processor.

[0038] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a load forecasting method under multiple extreme weather scenarios taking error correction into account.

[0039] Beneficial effects of the present invention: The load forecasting method under multiple extreme weather scenarios taking into account error correction provided by the present invention makes the forecasting results more accurate by collecting meteorological indicators to define extreme weather indicators and extreme weather classification. The analysis and screening links use dual-channel parallelism so that the present invention can obtain extreme weather types by inputting meteorological indicators, and can also obtain meteorological indicator combinations by inputting weather types, and then match the extreme weather corresponding to the indicator combination to complete the forecast of extreme weather. The present invention achieves better results in extreme weather forecasting, correction of forecast errors and parameter updating. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0041] Figure 1 An overall flow chart of a load forecasting system under multiple extreme weather scenarios taking error correction into account provided by the first embodiment of the present invention.

[0042] Figure 2 The first embodiment of the present invention provides an error correction process based on Bayesian hyperparameter optimization for a load forecasting method under multiple extreme weather scenarios taking error correction into account. DETAILED DESCRIPTION

[0043] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0044] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a preferred solution of a load forecasting method under multiple extreme weather scenarios taking into account error correction, including:

[0045] S1: The correlation between extreme weather classification and the meteorological indicators is analyzed through the Pearson correlation coefficient.

[0046] Furthermore, by screening the meteorological characteristic factors that affect extreme weather, analyzing the relationship between specific weather classifications and meteorological indicators, the meteorological indicators of weather classifications and meteorological indicators are obtained. The Pearson correlation coefficient is a statistical tool to measure the degree of linear correlation between two variables. Its value is between -1 and 1. The higher the absolute value of the correlation coefficient, the stronger the linear relationship between the variables. Its mathematical formula is shown in formula (1). In the impact of extreme weather on load forecasting, there are complex interactions between different meteorological factors, which jointly affect the accuracy of load forecasting. By calculating the Pearson correlation coefficient, the correlation between different meteorological factors and power load is quantitatively analyzed, so as to determine the characteristics of key meteorological factors affecting extreme weather. Table 1 gives the main reference characteristics of different extreme weather.

[0047] Table 1 Reference characteristics of main meteorological factors affecting extreme weather

[0048]

[0049]

[0050] It should be noted that in formula (1), X i and Y i are the i-th observation values ​​of variables X and Y, respectively, μ X and μ Y are the overall means of variables X and Y, respectively, and n is the number of observations. Based on the calculation results of the Pearson correlation coefficient, the coefficient distribution between 0.5-0.9 indicates that the correlation between two variables is medium-strong correlation. The nine parameters of wind speed, precipitation, precipitation intensity, duration, minimum temperature, maximum temperature, air pressure, density, and humidity are determined as the key meteorological factor characteristics.

[0051]

[0052] S2: Based on the meteorological indicators, comprehensively evaluate extreme weather to obtain extreme weather classifications corresponding to the meteorological indicators; based on the artificial testing instructions, simulate the meteorological indicators of the extreme weather classifications to obtain a simulated data set.

[0053] Furthermore, by constructing different weights and input parameters, a comprehensive evaluation index model for extreme weather is constructed; through the comprehensive evaluation index model for extreme weather, the specific functional relationship between extreme weather and meteorological indicators is reflected.

[0054] R H =κ h x1+κ y x2+...+κ d x n

[0055] In the first channel, meteorological indicators are input and brought into the comprehensive evaluation indicator model of each extreme weather classification through a matching mechanism to obtain an evaluation value for each extreme weather type. The specific extreme weather classification is matched through the correspondence between the evaluation value of each extreme weather and the threshold.

[0056] In the second channel, the extreme weather type is input, and a specific comprehensive evaluation index model is obtained based on the input extreme weather type. According to the threshold range of the evaluation value, the meteorological indicators in the comprehensive evaluation index model are assigned and randomly combined to obtain the meteorological indicators corresponding to the weather type.

[0057] Table 2 Comprehensive evaluation indicators of extreme weather

[0058]

[0059]

[0060] It should be noted that due to the significant differences in weather characteristics and environmental conditions in different regions, the definition of extreme weather may vary; therefore, in order to ensure the standardization and universality of the content studied in this paper, by combining the basic definition of extreme weather and the main indicators shown in Table 1 above, this paper constructs a comprehensive evaluation index for the definition of extreme weather, which can screen out extreme weather scenes within a certain time range according to a certain proportion. Taking high temperature heat waves, low temperature cold waves and heavy rains as examples, the comprehensive evaluation index of extreme weather is given as formula (2) which represents high temperature heat waves R HWI , Formula (3) represents the low temperature cold wave R CWI , Formula (4) represents the rainstorm R RSI .

[0061] R HWI =κ w W r +κ h H+κ b T b +κ y P (2)

[0062] Where W r is the temperature and the weight is κ w , H is the air humidity, and the weight is κ h , T b To represent the duration of high temperature, the weight is κ b , P represents the total precipitation within a certain period of time, and the weight is κ y κ w , κ h , κ b and κ y are all greater than 0 and the sum is 0, generally take κ y =0.4,κw =0.2,κ h =0.3,κ y =0.1.

[0063] R CWL =κ c ΔT+κ d T min +κ s T c (3)

[0064] Where ΔT represents the difference between the highest temperature and the lowest temperature, and the weight is κ c , T min represents the daily minimum temperature, with a weight of κ d , T c Indicates the duration of low temperature, with a weight of κ s κ c , κ d , κ s are all greater than 0 and the sum is 0, generally take κ c =0.4,κ d =0.4,κ s =0.2.

[0065] R RSI =κ y P+κ q I Q +κ r T R (4)

[0066] In the formula, P represents the total precipitation within a certain period of time, and the weight is κ y , I Q represents the amount of precipitation per unit time, and the weight is κ q , T R represents the duration of precipitation, with a weight of κ r ; The three weight values ​​are all greater than 0 and equal to 0, generally taking κ y =0.5,κ q =0.3,κ r =0.2.

[0067] S3: Use two channels to obtain initial parameters for load forecasting; collect clear meteorological indicators through channel one, and obtain artificial test instruction simulation through channel two.

[0068] Furthermore, the manual test instructions of channel two simulate the load forecasting method under multiple extreme weather scenarios taking into account error correction, including inputting a specific weather classification, obtaining the meteorological index corresponding to the input weather classification according to the Pearson correlation coefficient, and combining the meteorological indexes corresponding to the weather classification according to the calculation formula of the comprehensive evaluation index of extreme weather to simulate various extreme weather conditions.

[0069] If channel one obtains historical data, it is used to train the model; if it obtains future weather data predicted by the meteorological station, the model is used to calculate the load value of the future day.

[0070] Channel 2 obtains all artificial simulated meteorological data, which is used to characterize the load value under such climate characteristics on a certain day in the future.

[0071] It should be noted that the Pearson coefficient was used in the screening process of channel 2 to screen relevant meteorological factors with correlation coefficients between 0.5 and 0.9, and a comprehensive evaluation index of extreme weather was constructed to distinguish between normal weather and extreme weather events, providing a clear standard for data classification.

[0072] S4: Taking extreme weather classification and meteorological indicators as input, the temporal features of the time series are extracted, denoised, and reconstructed.

[0073] Furthermore, the ICEEMDAN algorithm is used to decompose the original multivariate time series load data.

[0074] The algorithm ICEEMDAN is used to decompose complex time series data into a series of intrinsic mode functions with different time scales.

[0075] The K-means clustering algorithm is used to classify the intrinsic mode function for calculating the sample entropy; four steps are respectively represented: distance measurement, cluster allocation, cluster center update, and objective function.

[0076] According to the complexity of the intrinsic mode function, high-frequency components are extracted from the intrinsic mode function.

[0077] The variational mode decomposition algorithm is applied to the eigenmode function belonging to the high frequency part for secondary decomposition; the signal is decomposed into the eigenmode function with specific frequency characteristics using the VMD calculation.

[0078] A threshold is selected according to a wavelet threshold denoising method; and components identified as containing noise are processed using the wavelet packet decomposition, threshold quantization and signal reconstruction techniques.

[0079] The ICEEMDAN initial decomposition stage uses the ICEEMDAN algorithm to decompose complex time series data into intrinsic mode functions with different time scales. After obtaining the intrinsic mode functions, the sample entropy of each intrinsic mode function is calculated to quantify its complexity; sample entropy is a statistic that measures the complexity and regularity of time series. It can reflect the randomness and uncertainty of the signal. Through the calculation of sample entropy, the complexity of each intrinsic mode function can be evaluated, providing a basis for subsequent clustering analysis.

[0080] The K-means clustering step uses the K-means clustering algorithm to classify the intrinsic mode functions for which sample entropy has been calculated; the intrinsic mode functions are grouped according to their frequency characteristics to ensure that the intrinsic mode functions in each category have similar frequency characteristics. This step helps to further simplify the data structure and provide clearer input features for subsequent prediction models.

[0081] After clustering processing, the high-frequency part of the intrinsic mode function is extracted and secondary decomposition is performed using variational mode decomposition; the variational mode decomposition algorithm is an adaptive decomposition method, which determines the optimal number of modes by constructing a variational constraint problem and accurately adjusts the center frequency and bandwidth of each mode. This algorithm can further decompose the high-frequency intrinsic mode function into an intrinsic mode sequence with clearer seasonal characteristics.

[0082] The sample entropy of the processed intrinsic mode sequence is calculated again, and the noise signal is eliminated based on the wavelet threshold denoising method. The principle is that the wavelet coefficients of the signal after wavelet transformation contain important information. By selecting a suitable threshold, the larger coefficients are retained and the smaller ones are set to 0, thereby retaining the signal and filtering out the noise.

[0083] It should be noted that the clear time series features are finally recombined to reconstruct the load time series; this step helps to reduce the complexity and non-stationarity of the input signal and provide a more accurate data basis for subsequent analysis and prediction.

[0084] Reference Figure 2 , which is the first embodiment of the present invention, provides an error correction method for a load forecasting method under multiple extreme weather scenarios taking error correction into account, comprising:

[0085] S5: Input the results of time series feature extraction into the load main prediction model and the error correction model to obtain the power load prediction results.

[0086] Furthermore, the subsequences are combined and reconstructed through the clear time series features after process processing; the sequence data is processed by two gated recurrent units, so that the bidirectional gated recurrent unit neural network model can simultaneously obtain the contextual information of each time point, and predict the changes in power load under extreme weather conditions that are jointly affected by historical and future trends.

[0087] The normalized exponential function is used to further improve the model prediction effect. The prediction error objective function f is used to evaluate the performance of the model, and the model parameters are optimized by minimizing this function.

[0088] Due to the time series characteristics and nonlinear features, as well as the complexity of hyperparameter selection of the CNN-BiGRU model, the Bayesian optimization method is used to dynamically adjust the search strategy for the existing evaluation results, identify the hyperparameter configuration, avoid falling into the local optimal solution, and achieve error correction.

[0089] Advanced deep learning technology is used to analyze and predict the multivariate time series decomposed and reconstructed in the first stage. The core of this stage is a CNN-BiGRU hybrid neural network model, which integrates the advantages of convolutional neural network (CNN) and bidirectional gated recurrent unit (BiGRU) to achieve in-depth learning and accurate prediction of load data. On this basis, a prediction error objective function is established, and the Bayesian method is used to optimize the hyperparameters of BiGRU, and the error of the model under multi-sample data training is continuously corrected to enhance the model's ability to cope with the mutation and strong randomness of load time series under extreme weather conditions.

[0090] The network structure of the CNN part consists of multiple convolutional layers and pooling layers. The convolutional layers can automatically learn important features of input data (such as historical load data and meteorological data), such as local patterns and trends. Through convolution operations, CNN can capture complex patterns and trends in power load data, especially under extreme weather conditions. The pooling layer is used to reduce the spatial dimension of the features while maintaining the main information of the features, which helps to reduce the complexity of subsequent calculations.

[0091] The BiGRU part is a bidirectional gated recurrent unit network that processes sequence data through GRU units in two directions. This structure enables BiGRU to simultaneously obtain the contextual information of each time point, which is particularly important for predicting changes in power load under extreme weather conditions that are affected by historical and future trends. The bidirectional nature of BiGRU enables it to better understand and predict long-term dependencies in time series data.

[0092] In order to further improve the model's attention to key information, a self-attention mechanism is introduced in the CNN-BiGRU model. This mechanism calculates the similarity scores between the query vector and all key vectors, and then uses the softmax function to convert these scores into normalized weights. Based on these weights, the model can perform weighted summation on the value vectors to generate a context vector, allowing the model to focus on the most important parts of the input data. This process helps the model identify and weight the features that have the greatest impact on the prediction results, thereby improving the accuracy and robustness of the prediction.

[0093] In order to improve the model's ability to cope with the mutation and strong randomness of time series under extreme weather, the present invention establishes a prediction error objective function and optimizes the hyperparameters of the BiGRU model through the Bayesian method; in the optimization process, the objective function, that is, the loss function of the model, such as the mean square error (MSE) or the mean absolute error (MAE), is defined, and it is used as the target to be minimized in the Bayesian optimization process; then, the search space of the hyperparameters is set, which includes the key parameters in the BiGRU model, such as the number of GRU units, batch size, learning rate, etc.; the Bayesian optimization gradually narrows the search range through an iterative process, and selects the hyperparameter combination that is most likely to improve the model performance for evaluation in each step; after the hyperparameters of the BiGRU model are optimized using the Bayesian method, the model is retrained using the optimized parameters to improve the performance of the model in the face of the mutation and strong randomness of time series data.

[0094] The CNN-BiGRU model combines the features of CNN and BiGRU. The CNN part is responsible for extracting the spatiotemporal features in the power load data, which include the local and global features of the power load. Through the convolutional layer and the pooling layer, the model can capture the complex patterns in the data and provide rich input information for subsequent predictions. The BiGRU part is responsible for processing sequence data, which can capture the long-term dependencies in time series data. The CNN-BiGRU model can not only process and integrate multiple data sources, such as historical load data, meteorological data (temperature, humidity, wind speed, etc.), holiday information, etc., but also adapt to different types of load data and meteorological variables, which enables the model to maintain high prediction accuracy and robustness when facing extreme weather conditions.

[0095] The CNN structure is usually composed of convolutional layers, activation functions, pooling layers, and fully connected layers; its convolutional layers can automatically learn important features of input data (such as historical load data and meteorological data). This feature extraction capability enables CNN to capture complex patterns and trends in power load data, especially under extreme weather conditions. The following equation (9) is the mathematical formula of CNN.

[0096]

[0097] In the above formula, l is the convolutional layer of the lth layer, is the output of the lth layer and the input of the l+1th layer, is the weight matrix, is the multiplicative bias and additive bias of the jth neuron in the lth layer, f() is the activation function used by the convolutional layer, and m j is the jth convolution region of the l-1 layer feature, and down() is the adopted function. is x i The exponential function of BiGRU is a bidirectional gated recurrent unit neural network model. It processes sequence data through two GRU units, which enables BiGRU to simultaneously obtain the context information of each time point. It is particularly important for predicting the changes in power load under extreme weather conditions that are jointly affected by historical and future trends. For an input of X = x1, x2, ... x T The mathematical model of BiGRU can be expressed as shown in formula (9).

[0098]

[0099] In the above formula, are the hidden states from left to right and from right to left, respectively, x t is the tth element of the input sequence, h t is the final hidden state, y t is the output, W and b are the weight and bias of the fully connected layer respectively; the CNN-BiGRU model combines the characteristics of CNN and BiGRU, aims to reduce the prediction error, and uses the BO algorithm to optimize the hyperparameters of the BiGRU network; to address the problem of high dimension of time series features, CNN is first used to extract the features of the historical load time series, and then BiGRU is used to process the feature vectors of the historical load series. Finally, Attention is used to calculate the different weights in the fully connected layer to obtain the short-term load forecasting results.

[0100] Attention is a technique used in deep learning to enhance the model's attention to important parts of the input data. Introducing this technique in the CNN-BiGRU model can further enhance the model's attention to key information, allowing the model to pay more attention to the time steps that contribute to the prediction results, thereby improving the accuracy and robustness of the prediction. The process is to calculate the similarity scores between the query vector and all key vectors, and then use the softmax function to convert these scores into normalized weights. Finally, the value vectors are weighted and summed according to these weights to generate a context vector, so that the model can focus on the most important part of the input data. Formula (10) represents the calculation process of the Attention mechanism.

[0101]

[0102] In formula (11), e t Q t The similarity score between and K, α t is the normalized weight, indicating Q t The corresponding value vector V i The importance of V is determined by using the softmax function to ensure that the sum of all weights is 1; i is the value vector at time step i, and the context vector c t Used for prediction of the next step or as input to the GRU layer.

[0103] Prediction error objective function, which is used to calculate the difference between the model's predicted value and the actual value. This function is the key indicator that needs to be minimized during the model optimization process. The hyperparameters of the BiGRU model are adjusted using the Bayesian optimization method; Bayesian optimization is a global optimization strategy based on Bayesian statistical inference. It constructs a probability model of the hyperparameter space and uses prior knowledge and historical evaluation information to guide the search for the optimal hyperparameters.

[0104] It should be noted that the present invention proposes a prediction error objective function f, which is a measure of the model prediction value y i ′ and the actual observed value The difference between i The mathematical expression of ; in the prediction task, the error function is used to evaluate the performance of the model and optimize the model parameters by minimizing this function. Specifically, the error function can be defined as a quantitative measure of the deviation between the predicted value and the actual value. and the actual load value y i s Input the model and get the predicted value y i ′; the internal parameters of the model are That is, when each set of training data is input, the parameter value set generated is used to find the optimal model parameters using Bayesian hyperparameter optimization to minimize the error and, when the first set of data enters the model, the parameter set is obtained. After the second set of data enters the model, the parameter set is obtained Using Bayesian hyperparameter optimization model to meet f B After that, we get the updated parameter set The above process is repeated until all i groups of training data are used and the model parameters are updated i-1 times. The principle is shown in the following formula.

[0105]

[0106] In the above formula, f Ais the final error target obtained in the first input model, which has not been updated yet; the final The prediction error objective function f is minimized, and the error between the predicted value and the actual value will be very small when the model parameters after training are updated to predict the test set, thus achieving accurate load forecasting.

[0107] In view of the time series characteristics and nonlinear features of power load data, as well as the complexity of hyperparameter selection of the CNN-BiGRU model, this paper adopts the Bayesian optimization method to dynamically adjust the search strategy according to the existing evaluation results, identify the better hyperparameter configuration, and avoid falling into the local optimal solution to achieve error correction; the core mechanism of BO is to update the posterior probability distribution of the objective function based on the Bayesian theorem, and intelligently select new hyperparameter combinations for testing based on this distribution. By continuously iteratively updating the prior knowledge, the globally optimal hyperparameter setting is determined, as shown in the following formula (13), which is the Bayesian optimization expression.

[0108]

[0109] Where: f(h) represents the prior distribution model, h * represents the optimal parameter value of the f(h) constraint domain, such as activation function, learning rate, and convolution kernel size of CNN convolution layer, etc. H represents the candidate set. The Bayesian optimization algorithm has the advantage of being able to process multiple variables at the same time. The specific process is as follows Figure 2 shown.

[0110] Example 2 is an embodiment of the present invention, which provides a load forecasting method under multiple extreme weather scenarios taking into account error correction. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0111] First of all, in order to verify the effectiveness and accuracy of the method proposed in this paper, three groups of control algorithm models, namely CNN-LSTM (CL), CNN-BiGRU (CB) and ICEEMADAN-CNN-BiGRU (ICB), were constructed with reference to relevant literature, and the same data set was trained. The same prediction data set was then used for comparative analysis with the algorithm ICEEMADAN-Kmeans-VMD-CNN-BiGRU-Attention (IKVBA) proposed in this paper.

[0112] In order to verify the effectiveness and superiority of the method proposed in this paper in dealing with extreme weather conditions, this paper selected the historical load data of the past three years in southwest China, including Sichuan, Chongqing, Guizhou, Yunnan and Tibet, for verification. Among them, the extreme weather: real data set contains 2047 sets of data. At the same time, due to the scarcity of extreme weather data, in order to fully train the data model, data enhancement technology is used. Through random perturbations or rule-based changes, 5453 sets of extreme weather and load data are created. Combined with the original data, a total of 7500 sets of data, covering a variety of extreme weather conditions, mainly high temperature weather data; normal weather is 7500 sets of data in the above five regions. In order to ensure that the model can effectively learn and generalize at different stages, these data are divided into normal and extreme weather categories and randomly divided into training sets, validation sets and test sets in a ratio of 7:2:1.

[0113] In order to verify the effectiveness and accuracy of the method proposed in this paper, three groups of control algorithm models, namely CNN-LSTM (CL), CNN-BiGRU (CB) and ICEEMADAN-CNN-BiGRU (ICB), were constructed with reference to relevant literature, and the same data set was trained. The same prediction data set was used for comparative analysis with the algorithm ICEEMADAN-Kmeans-VMD-CNN-BiGRU-Attention (IKVBA) proposed in this paper. Table 3 shows the data prediction results for normal weather, and Table 4 shows the data prediction results for extreme weather.

[0114] Table 3 Comparison of normal weather prediction performance of four models

[0115]

[0116] As shown in Table 3, the prediction accuracy of CL, CB, ICB and IKVBA is gradually improved, and the performance of IKVBA model is significantly better than the other three models. The average values ​​of MAPE, MAE, RMSE and R2 of CL, CB and ICB models are 2.75%, 0.247, 0.293 and 92.64% respectively. The IKVBA model improves the average values ​​of the other three models by 43.02%, 44.25%, 35.80% and 7.07% respectively in the four indicators, and improves the indicators of ICB by 24.05%, 29.24%, 16.20% and 3.72% respectively.

[0117] Table 4 Comparison of extreme weather prediction performance of four models

[0118]

[0119]

[0120] As shown in Table 4, the average values ​​of MAPE, MAE, RMSE and R2 of the three models of CL, CB and ICB are 5.85%, 0.444, 0.62 and 77.76% respectively. The performance of the IKVBA model is significantly better than that of the other three models.

[0121] Through these two sets of data, we can clearly see the prediction performance advantages of the IKVBA model under normal and extreme weather conditions, especially in terms of accuracy and reliability. Especially the good advantages in extreme conditions prove the practicality of error correction.

[0122] Through these two sets of data, it can be clearly seen that the prediction performance advantages of the IKVBA model under normal and extreme weather conditions, especially in terms of accuracy and reliability. Especially the good advantages in extreme cases prove the practicality of error correction. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those of ordinary skill in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0123] Embodiment 3 is an embodiment of the present invention, which provides a load forecasting system under multiple extreme weather scenarios taking into account error correction, including a classification module, an analysis and simulation module, and a prediction module.

[0124] The classification module is used to: analyze the correlation between extreme weather classification and the meteorological index through the Pearson correlation coefficient; obtain the extreme weather classification corresponding to the meteorological index according to the meteorological index, and establish a comprehensive evaluation system for extreme weather taking into account the meteorological index.

[0125] The analysis and simulation module is used to: use two channels to obtain initial parameters for load forecasting; collect clear historical meteorological data sets through channel one, and obtain artificial test simulation meteorological data sets through channel two.

[0126] The prediction module is used to: take the classification of extreme weather and meteorological index data sets as input, decompose, denoise and reconstruct the temporal characteristics of the time series; input the results of the temporal characteristics extraction into the load main prediction model and the error correction model to obtain the power load prediction results.

[0127] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0128] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0129] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0130] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0131] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A load forecasting method under multiple extreme weather scenarios taking into account error correction, characterized in that: include: The correlation between extreme weather classification and meteorological indicators was analyzed through the Pearson correlation coefficient; According to the meteorological index, an extreme weather classification corresponding to the meteorological index is obtained, and a comprehensive extreme weather evaluation system considering the meteorological index is established; Using two channels, the initial parameters for load forecasting are obtained; Channel 1 is used to collect clear historical meteorological data sets, and channel 2 is used to obtain artificial test simulated meteorological data sets; Taking extreme weather classification and meteorological indicator datasets as input, the temporal features of the time series are decomposed, denoised, and reconstructed; The results of time series feature extraction are input into the main load prediction model and error correction model to obtain the power load prediction results.

2. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 1, characterized in that: The Pearson correlation coefficient includes obtaining the main meteorological indicators that affect each extreme weather by screening the meteorological characteristic factors that affect extreme weather and analyzing the relationship between specific weather classifications and meteorological indicators.

3. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 2, characterized in that: The comprehensive evaluation includes constructing an extreme weather comprehensive evaluation index model by constructing different weights and input parameters; R H =k h x1+k y x2+...+k d x n Among them, R H is the evaluation value of extreme weather type, x1, x2...x n is the meteorological index of extreme weather type, and the meteorological index weight is κ h , κ d ... y , are all greater than 0 and the sum is 0; The specific functional relationship between extreme weather and meteorological indicators is reflected through the extreme weather comprehensive evaluation index model.

4. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 2, characterized in that: The artificial test instruction simulation includes inputting a specific weather classification, obtaining the meteorological index corresponding to the input weather classification according to the Pearson correlation coefficient, combining the meteorological indexes corresponding to the weather classification according to the calculation formula of the extreme weather comprehensive evaluation index, and simulating various extreme weather conditions; In the first channel, meteorological indicators are input and brought into the comprehensive evaluation indicator model of each extreme weather classification through the matching mechanism to obtain the evaluation value for each extreme weather type. Through the correspondence between the evaluation value of each extreme weather and the threshold, the specific extreme weather classification is matched; In the second channel, the extreme weather type is input, and a specific comprehensive evaluation index model is obtained based on the input extreme weather type. According to the threshold range of the evaluation value, the meteorological indicators in the comprehensive evaluation index model are assigned and randomly combined to obtain the meteorological indicators corresponding to the weather type.

5. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 3, characterized in that: The load time series includes performing secondary decomposition and reconstruction on the load time series to form a time series with high noise removal and clear features; The specific steps are: The ICEEMDAN algorithm is used to decompose the original multivariate time series load data; The algorithm ICEEMDAN is used to decompose complex time series data into a series of intrinsic mode functions with different time scales; The K-means clustering algorithm is used to classify the intrinsic mode function of the sample entropy; the four steps of distance measurement, cluster allocation, cluster center update, and objective function are respectively represented; Extracting high-frequency components of the intrinsic mode function according to its complexity; Applying variational mode decomposition algorithm to perform secondary decomposition on the eigenmode function belonging to the high frequency part; using the calculation of variational mode decomposition algorithm to decompose the signal into eigenmode functions with specific frequency characteristics; A threshold is selected according to a wavelet threshold denoising method; and components identified as containing noise are processed using the wavelet packet decomposition, threshold quantization and signal reconstruction techniques.

6. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 5, characterized in that: The decomposition, denoising and reconstruction include combining and reconstructing subsequences through clear time series features after process processing; By processing sequence data through two gated recurrent units, the bidirectional gated recurrent unit neural network model can simultaneously obtain the contextual information of each time point and predict the changes in power load under extreme weather conditions that are jointly influenced by historical and future trends.

7. The load forecasting method under multiple extreme weather scenarios taking into account error correction as claimed in claim 6, characterized in that: The input load main prediction model and error correction model include using a normalized exponential function to further improve the model prediction effect; the prediction error objective function f is used to evaluate the performance of the model, and the model parameters are optimized by minimizing this function; Due to the time series characteristics and nonlinear features, as well as the complexity of hyperparameter selection of the CNN-BiGRU model, the Bayesian optimization method is used to dynamically adjust the search strategy for the existing evaluation results, identify the hyperparameter configuration, avoid falling into the local optimal solution, and achieve error correction.

8. A power load forecasting system under multiple extreme weather scenarios taking into account error correction according to any one of claims 1 to 7, characterized in that: Including classification module, analysis and simulation module, and prediction module; The classification module is used to: analyze the correlation between extreme weather classification and the meteorological index through the Pearson correlation coefficient; obtain the extreme weather classification corresponding to the meteorological index according to the meteorological index, and establish a comprehensive evaluation system for extreme weather taking into account the meteorological index; The analysis and simulation module is used to: use two channels to obtain initial parameters for load forecasting; collect clear historical meteorological data sets through channel one, and obtain artificial test simulation meteorological data sets through channel two; The prediction module is used to: take the classification of extreme weather and meteorological index data sets as input, decompose, denoise and reconstruct the temporal characteristics of the time series; input the results of the temporal characteristics extraction into the load main prediction model and the error correction model to obtain the power load prediction results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images